ARMADA: Using motif activity dynamics to infer gene regulatory networks from gene expression data

Pemberton-Ross, Peter; Nimwegen, Erik van

doi:10.1016/j.ymeth.2015.06.024

Cited by 5 publications

(4 citation statements)

References 65 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this vein, foundation models remain largely unexplored, and we specifically note that our feature extraction strategy is not the only way GeneFormer could have been used. Another area with promising individual demonstrations is expression forecasting from time-series or RNA velocity data (Burdziak et al, 2023;Kamimoto et al, 2023;Pemberton-Ross et al, 2015;Qiu et al, 2022;Yeo et al, 2021). However, in light of our results, claims of causal identification from transfer learning across cell types or from time-series data should be tested by thorough and neutral benchmarks that include maximally informative baseline methods.…”

Section: Discussionmentioning

confidence: 94%

“…Expression forecasting methods also vary widely in terms of implementation and expected input. Certain methods use prior drafts of causal network structure (CellOracle, Dictys, ScanBMA, scREMOTE, scKINETICS, D-SPIN); prior knowledge of gene-gene functional relatedness (GEARS); or pre-training on millions of cells (scGPT, GeneFormer) (Burdziak et al, 2023;Cui et al, 2023;Jiang et al, 2023;Kamimoto et al, 2023;Lopez, Hütter, et al, 2022;Pemberton-Ross, Pachkov, & van Nimwegen, 2015;Qiu et al, 2022;Roohani et al, 2022;Theodoris et al, 2023;Wang et al, 2022;Yeo et al, 2021;Young, Raftery, & Yeung, 2014). Regarding analytical choices, key distinctive features include different regression methods (DCD-FG, CellOracle, Dynamo), use of low-rank structure (DCD-FG, ARMADA, D-SPIN); special handling of complex time-series (PRESCIENT); and modeling of transcription rates in addition to transcript levels (Dynamo, scKINETICS).…”

Section: Abstract Introduction Resultsmentioning

confidence: 99%

“…Suspecting overfitting or lack of causal idenfitication, we sought to reduce the sample complexity of this task. Many methods restrict causal relationships by predicting a target gene's expression using only TF's with nearby motif occurrences in promoters or enhancers (Kamimoto et al, 2023;Pemberton-Ross et al, 2015;Wang et al, 2022;Young et al, 2014), yielding regression problems with fewer predictors and therefore lower-variance estimates. We collected over 10 previously published networks constructed by coexpression analysis, Bayesian edge prediction, and motif analysis of candidate cis-regulatory elements (Table 4).…”

Section: Resultsmentioning

confidence: 99%

See 2 more Smart Citations

A systematic comparison of computational methods for expression forecasting

Kernfeld,

Yang,

Weinstock

et al. 2023

Preprint

View full text Add to dashboard Cite

Due to the abundance of single cell RNA-seq data, a number of methods for predicting expression after perturbation have recently been published. Expression prediction methods are enticing because they promise to answer pressing questions in fields ranging from developmental genetics to cell fate engineering and because they are faster, cheaper, and higher-throughput than their experimental counterparts. However, the absolute and relative accuracy of these methods is poorly characterized, limiting their informed use, their improvement, and the interpretation of their predictions. To address these issues, we created a benchmarking platform that combines a panel of large-scale perturbation datasets with an expression forecasting software engine that encompasses or interfaces to current methods. We used our platform to systematically assess methods, parameters, and sources of auxiliary data. We found that uninformed baseline predictions, which were not always included in prior evaluations, yielded the same or better mean absolute error than benchmarked methods in all test cases. These results cast doubt on the ability of current expression forecasting methods to provide mechanistic insights or to rank hypotheses for experimental follow-up. However, given the rapid pace of innovation in the field, new approaches may yield more accurate expression predictions. Our platform will serve as a neutral benchmark to improve methods and to identify contexts in which expression prediction can succeed.

show abstract

Section: Discussionmentioning

confidence: 94%

Section: Abstract Introduction Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

A systematic comparison of computational methods for expression forecasting

Kernfeld,

Yang,

Weinstock

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Furthermore, modern multi-omic methods merge mRNA measurements with much more molecular information, and this may suffice to capture influences missed in mRNA data. In particular, genome-wide averages of activity near specific motifs may contain information about TF activity that is not present in transcript counts (Pemberton-Ross, Pachkov, & van Nimwegen, 2015). To evaluate knockoff filter FDR control on multi-omic data, we turned to a mouse skin and hair follicle dataset consisting of paired RNA and chromatin measurements on 34,774 single cells from female mice (S. Ma et al, 2020).…”

Section: Conditional Dependence Does Not Imply Direct Regulation In M...mentioning

confidence: 99%

Model-X knockoffs reveal data-dependent limits on regulatory network identification

Kernfeld

Keener

Cahan

et al. 2023

Preprint

View full text Add to dashboard Cite

Computational biologists have long sought to automatically infer transcriptional regulatory networks (TRNs) from gene expression data, but such approaches notoriously suffer from false positives. Two points of failure could yield false positives: faulty hypothesis testing, or erroneous assumption of a classic criterion called causal sufficiency. We show that a recent statistical development, model-X knockoffs, can effectively control false positives in tests of conditional independence in mouse and E. coli data, which rules out faulty hypothesis tests. Yet, benchmarking against ChIP and other gold standards reveals highly inflated false discovery rates. This identifies the causal sufficiency assumption as a key limiting factor in TRN inference.

show abstract